水体富营养化状况的人工神经网络预测模型的建立

目的建立郑州市主要生活饮用水源富营养化预测模型。方法以郑州市主要生活饮用水源西流湖和黄河花园口段某调蓄池作为研究现场,监测富营养化相关理化指标水温(WT)、透明度(SD)、总磷(TP)、总氮(TN)、光照度(Li)、高锰酸盐指数(CODMn)、叶绿素a(Chla);采用评分法和综合营养状态指数法对水体富营养化状况进行评价;运用标准化函数对各理化因子原始监测数据进行归一化处理后,构建BP人工神经网络富营养化预测模型;并采用改进的Levenberg-Marquardt算法对网络进行优化。结果两水源富营养化状况评价结果显示:西流湖和黄河花园口段某调蓄池水体均为富营养化状态,西流湖随着春、夏、秋季节的变化,富营养化状况逐步加剧;应用BP神经网络技术,根据J=n+m+a函数计算隐层数,将隐层节点数范围定位2～15,分别进行训练,最终确定网络节点数为10;根据郑州市主要生活饮用水源环境理化指标检测数据,建立了水体富营养化相关因子Chla预测模型,重要的6个富营养化理化因子全部纳入模型进行分析,网络训练过程均方差为1e-11,实测值与预测值模型拟合相关系数为0.871,与预期目标比较接近,成功构建了郑州市主要生活饮用水源富营养化人工神经网络预测模型。结论人工神经网络技术能够运用于水体富营养化预警系统的研究,所建立预测模型更符合实际情况。

Establishment of the predictive model of source eutrophication using artificial neural network

Objective To establish the predictive model of eutrophication in the main water source in Zhengzhou City.Methods Water temperature(WT),secchi-depth(SD),total phosphorus(TP),total nitrogen(TN),light illuminance(LI),chemical oxygen demand(CODMn),chlorophyll-a(Chla) were monitored in Xiliu lake and Huayuankou pool.Grading points method and comprehensive trophic state index method were used to evaluate the trophic state.Backpropagation artificial neural network with Levenberg-Marquardt algorithm was used to establish the forcasting model of eutrophication after the raw data normalized treated using standardization function.Results The results of evaluation of grade method revealed that the two waters source were in nutritional state and the tendencies of year grade indexes were from the lower critical value to eutrophic state to higher critical value of eutrophication of xiliu lake.The scope of hidden nodes was determined from 2 to 15 according to the calculated results using function of J=n+m+a and the hidden nodes was 10 according to the training result.All of the physical chemistry factors were brought into the model.The training error was 1e-11 and the coefficient correlation of the network fitness result was 0.871.The fitting result was close to the aim and the predictive model of eutrophication in the main resource water in Zhengzhou City was established successfully.Conclusion Eutrophication forcasting model could be established using artificial neural network,and the method of artificial neural network should be better to meet the actual demand.